- Shares drift lower after keynote speech
- CEO unable to stem tide of selling
- New products shipping next quarter
Despite promising devotees at a packed hockey stadium in San Jose that Nvidia’s (NVDA:NASDAQ) AI chips would soon be powering self-driving cars and futuristic robots, chief executive Jensen Huang seems to have lost his touch.
Huang’s keynote address at the firm’s second GPU Technology Conference last night failed to lift Nvidia’s shares, which drifted $4.10 or 3.5% lower to $115.43 having already lost 14% so far this year.
BIG SPENDERS
Whether the lack of enthusiasm was down to the general ‘risk-off’ mood among investors, fears that spending by the ‘hyperscalers’ will be cut back in the event the US economy goes into recession or just a sense that Huang has hyped his company’s products a little too much and is in danger of over-promising, is hard to say.
Nvidia’s GTC, which brings together global developers, chip users and investors, is a key event in the calendar and the perfect forum for Huang to convince the likes of Alphabet (GOOG:NASDAQ), Amazon (AMZN:NASDAQ) and Meta Platforms (META:NASDAQ) that they should keep buying the firm’s products.
According to Bloomberg, spending on AI infrastructure is predicted to top $370 billion this year rising to more than $500 billion annually by 2032.
However, since the emergence of Chinese start-up DeepSeek’s low-cost generative AI model in January this year there have been questions over the need to spend hundreds of billions of dollars training new large-language models.
NEW PRODUCTS
Huang also used the GTC to unveil the Blackwell ‘Ultra’ supercomputing chip which the firm is due to begin shipping to customers in the second quarter and which it hopes will cement its dominance in AI.
The ‘Ultra’ can handle more tasks than first-generation Blackwell chips, with claims it could allow cloud providers to generate 50 times as much revenue per unit.
‘We designed Blackwell Ultra for this moment — it’s a single versatile platform that can easily and efficiently do pretraining, post-training and reasoning AI inference,’ said Huang.
Meanwhile, Huang admitted a new power-saving technology which uses laser beams to send data between chips instead of copper fibres and aims to cut energy use isn’t yet reliable enough for the company to use in its flagship chips.
These ‘switch chips’ will come out in small numbers later in 2025 but there are no plans to integrate them with Nvidia’s best-selling processors because copper connectors are ‘orders of magnitude’ more reliable, the chief executive told reporters.
